Page 3 of 4 FirstFirst 1234 LastLast
Results 31 to 45 of 46

Thread: 2.5 Ghz G5 Disproportionately Faster in LW 8?

  1. #31
    PC-killer Ge4-ce's Avatar
    Join Date
    Apr 2003
    Location
    Belgium
    Posts
    675
    Originally posted by harlan
    I'm sorry, but that just cracked me up!!
    I don't quite follow you.. have I said something wrong?

    If I did, what I meant was: I'm sceptical about marketing from Apple. I believe when tests done by users and tests done by myself are hitting the net..
    __________________________________________________ _



    Koen Delbroek
    D4motion
    www.d4motion.be
    [email protected]

  2. #32
    The Dude harlan's Avatar
    Join Date
    Feb 2003
    Location
    Caina, Cocytus
    Posts
    1,259
    No no...it was just funny, rather than say: I do not "trust" benchmarks. You must've miss typed and said: I do not "thrust" benchmarks.

    the misspelling may not be as funny across the language barrier.
    harley

    "it's better to live one day as a lion, than a thousand years as a lamb."

  3. #33
    PC-killer Ge4-ce's Avatar
    Join Date
    Apr 2003
    Location
    Belgium
    Posts
    675
    WOOOPs God**** typos... well.. if English isn't your native language, typos are a rule rather than exception..

    thanks to point me out on this one.. it IS actually quite funny that sentense got a whole other meaning like that

    Could have been worse though.. "I do not thrust benchmarks, I shake them by hand"
    __________________________________________________ _



    Koen Delbroek
    D4motion
    www.d4motion.be
    [email protected]

  4. #34
    Aha! A clue Sherlock! Jimzip's Avatar
    Join Date
    Feb 2003
    Location
    Melbourne Australia
    Posts
    399
    Originally posted by Lightwolf
    Well, let's try to verify the bench then.
    Anyone have the numbers for a Dual G5 2GHz?
    On my Xeon 3.06 it takes 90s with 8 threads.

    Cheers,
    Mike
    I'm getting a 2.0GHz early next week so I'll post some results for you.

    Now.
    Just to clarify a few things.
    Although the benchmarks look disproportionate, I believe them to be right. Not because I'm a zealot, but because I read up on the topic.
    Here's the information, what I've gathered & what I know:
    The 2.5GHz G5 is an improved system from even the 2.0GHz G5 models in many ways, not just because of the 500MHz ramp.
    Consider the fact that it has a larger frontside bus, that the processor is a 970fx, not a 970 (and believe me, there is a difference. It is a smaller unit, and utilises less power in general than the 970). Then of course, the cooling system. Which is said to be much more efficient than conventional fans, and fans used in the 1.8 & 2.0GHz models.
    Take a look at Tom's hardware guide and see what cooling the processor can do to speed (check out the overclocking to 5GHz video if you haven't seen it). Cooling achieves a heck of a lot. Now, let's go back to the original comparison.



    With the larger frontside bus, a ramp of .5GHz, a more effective & efficient cooling system, a smaller and more refined motherboard and a new revision processor, it's a lot easier to see why it is faster than even the next model down, I see these benches to be accurate.

    Make sense?

    Reading is fun...

    Jimzip
    Last edited by Jimzip; 06-25-2004 at 09:53 PM.
    You know you've been learning 3D for too long when you start seeing in procedurals...

  5. #35
    The cooling or smaller CPU Die does nothing to improve speed at the same clock frequency, all they do is allow higher freqencies.

    As toby already pointed out, you're all reading the chart wrong!
    the dual 2.5GHz is NOT 42% faster than the dual 2GHz, it is another 42% faster than the BASEILE, which is the 3.4GHz Pentium4.

    If you write the chart in total numbers it would read:

    Dual 2.5GHz G5: 204%
    Dual 2.0GHZ G5: 162%
    Dual 1.8GHz G5: 147%
    Dual 3.2GHz Xeon: 164%
    2.4GHz Opteron: 93%
    3.4 GHz Pentium4: 100% (Baseline)

    So 204%/162% (or 2.04/1.62) makes 1.259... or 126%, or 26% faster. At, exactly 25% more clockspeed.
    Last edited by Lynx3d; 06-25-2004 at 05:17 AM.

  6. #36
    Aha! A clue Sherlock! Jimzip's Avatar
    Join Date
    Feb 2003
    Location
    Melbourne Australia
    Posts
    399
    Originally posted by Lynx3d
    The cooling or smaller CPU Die does nothing to improve speed at the same clock frequency, all they do is allow higher freqencies.

    As toby already pointed out, you're all reading the chart wrong!
    the dual 2.5GHz is NOT 42% faster than the dual 2GHz, it is another 42% faster than the BASEILE, which is the 3.4GHz Pentium4.

    If you write the chart in total numbers it would read:

    Dual 2.5GHz G5: 204%
    Dual 2.0GHZ G5: 162%
    Dual 1.8GHz G5: 147%
    Dual 3.2GHz Xeon: 164%
    2.4GHz Opteron: 93%
    3.4 GHz Pentium4: 100% (Baseline)

    So 204%/162% (or 2.04/1.62) makes 1.259... or 126%, or 26% faster. At, exactly 25% more clockspeed.
    Oh yeah, maths was never my strong spot.. I was calculating it based on the 2.0GHz..
    Well there goes my logic..

    But can I just clarify something.
    Cooling a CPU improves electric conductivity with the metal in the transistors and allows more electrons to be passed through without jumps/bubbles which is faster, true?
    If that's true, and my logic is stemming from the fact that any CPU that is cooled will run faster, then shouldn't this be the same for the G5?
    Is this right?

    Jimzip
    Last edited by Jimzip; 06-25-2004 at 09:55 PM.
    You know you've been learning 3D for too long when you start seeing in procedurals...

  7. #37
    PC-killer Ge4-ce's Avatar
    Join Date
    Apr 2003
    Location
    Belgium
    Posts
    675
    Originally posted by Jimzip
    Oh yeah, maths was never my strong spot.. I was calculating it based on the 2.0GHz..
    Well there goes my logic..

    But can I just clarify something.
    Cooling a CPU improves electric conductivity with the metal in the transistors and allows more electrons to be passed through without jumps/bubbles which is faster, true?
    If that's true, and my logic is stemming from the fact that any CPU that is cooled will run faster, then shouldn't this be the same for the G5?
    Is this right?

    Jimzip
    I don't think that's the point here. I think that this G5 2.5 could be easily cooled down with traditional fans. Only they would produce more or a lot more noise than the current dual 2.0's..

    I think Apple just has chosen a different way to keep the machine quiet. And because liquid cooling is far more effecient, the fans that cool the hot liquid down have to be running at the same or even lower rpm as the previous models.

    I also don't think a cooled chip will automatically make the electrons go faster. In pure physics, if you would cool down a chip to -273,15°C or 0° Kelvin, the absolute minimum, you would have zero resistence, meaning the electrons could flow through without producing ANY heat. this on itselves would not produce a faster chip, but what it DOES allow is faster clockspeeds, and in that way make faster processors possible.

    just my 2ct
    Last edited by Ge4-ce; 06-26-2004 at 02:52 AM.
    __________________________________________________ _



    Koen Delbroek
    D4motion
    www.d4motion.be
    [email protected]

  8. #38
    Hehe, a lot of nice theories

    First, 2.5GHz are 2.5GHz. Full stop.
    If an individual transistor switches a lot faster this has no impact on performance because the result is taken every clock tick. Only if you want to increase clock speed you have to ensure the transistors will all have switched when you take the result (simplified explained...)
    To make it faster at the same clock speed, you have to improve the core logic (e.g. larger caches with lower latencies, better branch prediction etc.) But since the 970FX is nothing more than a 90nm version of the 970, it should perform exactly identical at the same frequency.

    Also, the resistance in the metal will probably be you smalles concern on a 90nm fabbed chip. I only had one semester about micro-electronic systems, but there are a LOT of effects changing with temperature. The most concerning seems to become leakage currents, which also increase with temperature. Silicon itself is not conducting at all (it is a semiconductor), it only conducts due to doping (fortunetaly!).
    The metal would perhaps be superconductive at 0° Kelvin (not sure about copper and aluminium used for interconnections), however the doped silicon will be useless because the dopands need some thermal energy to actually be "free" to do their work, so don't expect a chip to work at 0° Kelvin...but at too high temperature some parts may become more conducting than you want them to be.

  9. #39
    So 204%/162% (or 2.04/1.62) makes 1.259... or 126%, or 26% faster. At, exactly 25% more clockspeed

    So I was right! It's disproportionately faster!

  10. #40
    A few bits of information:

    The new dual 1.8 and dual 2.0 models still use the 'old' PPC970, manufactured with 130nm structure sizes; only the dual 2.5 model uses the new PPC970FX, manufactured with 90nm structure sizes. This has been verified by examinations of actual machines (using Apple's CHUD tools, one can read out the 'Processor Version Register' and identify the exact chip type.) So it is to be expected that the 'old' and 'new' 2.0GHz machines perform equally well.

    A disproportional increase in speed between the 2.0GHz and the 2.5GHz model is indeed an indicator that _something_ in the 2.5GHz models was improved beyond simply cranking up the clock frequency. It would be unusual even if the 2.5GHz machines gained 'only' the full 25%, because the chips surrounding the processor are not clocked faster, only the processor itself is.

    It is not clear whether the CPU itself is faster, it could also be improvements in the memory controller (but it is still unknown whether the 2.5GHz model has a new System Controller chip).

    Apple's benchmarks are probably real as far as comparisons between Mac models are concerned (they are the same platform, after all); it is most likely that the 2.5GHz's 'too good' performance is simply a sign of continued advancements of the processor and/or the surrounding system. MHz isn't the only thing that nets you more speed.

  11. #41
    obfuscated SDK hacker Lightwolf's Avatar
    Join Date
    Feb 2003
    Location
    Stuttgart, Germany
    Posts
    13,613
    Originally posted by hobold
    MHz isn't the only thing that nets you more speed.
    This is true for most situation, except for rendering, where the main loops tends to be in the CPU cache anyhow. Rendering benchmarks tend to scale linearily with CPU frequency on the same brand of CPU.
    On the other, the remaining 1% is well within any measuring tolerance, and could attributed to just about anything...

    Cheers,
    Mike

  12. #42
    Aha! A clue Sherlock! Jimzip's Avatar
    Join Date
    Feb 2003
    Location
    Melbourne Australia
    Posts
    399
    "So everything Tom told me was a lie?
    Everything!
    Why Tom?! WHY!?!?!" (last line delivered with yelling, fists raised and in the middle of a thunderstorm..)

    Jimzip
    Last edited by Jimzip; 06-26-2004 at 07:37 PM.
    You know you've been learning 3D for too long when you start seeing in procedurals...

  13. #43
    I haven't made quite clear what I meant. The move from 2.0GHz to 2.5GHz alone cannot explain the 26% performance gain. While there are some benchmarks that are only limited by core clock speed, rendering of any reasonably complex scenes will be impacted by memory speed as well. For one, 512KB of on-chip cache aren't enough to hold a scene. And secondly, memory accesses during rendering tend to have only medium 'locality of reference', mostly due to non-local stuff like shadows and reflections.

    Apple's benchmarked scene does scale fairly well, but the 11% clock speed increase from 1.8GHz to 2.0GHz only nets 10% performance increase. This means the 2.5GHz model is at least 3% 'too fast'.

    My comment 'more than just MHz' was meant to underscore that either the PPC970FX has gained aditional performance improvements over the PPC970 (even at the same clock speed), or that other important parts of the machine have been improved. It is more likely that the PPC970FX was improved, because system enhancements would have helped the 1.8 and 2.0GHz models, too.

    (This is not meant to invalidate the statement that CPU core speed is by far the most important factor of the overall performance of a renderer like Lightwave.)

  14. #44
    If you insist in 1 or 2 percent being a core improvement...that's a very lousy one

    if those number were 146,51%, 162,49% and 203,55% and they just rounded to whole percentages everything is within less than 0.5% linear...
    And the scene itself does not need to fit in cache to be considered memory independant. An Athlon with 512kb Cache is less than 1% faster in LW than one with 256kB cache IIRC...but it also depends on architecture, P4 is more sensitive to cache (or lack thereof)

  15. #45
    Originally posted by Lynx3d
    [B]If you insist in 1 or 2 percent being a core improvement...that's a very lousy one
    I am really thinking of bug fixes rather than true architectural improvements. The 970 was the first release of a new chip, so there were probably a few minor glitches. Some of the required workarounds could have cost a tiny bit of performance here and there. Those emergency brakes would now be unlocked in the 970FX.

Page 3 of 4 FirstFirst 1234 LastLast

Bookmarks

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •